CN107292307B - Automatic identification method and system for inverted Chinese character verification code - Google Patents
Automatic identification method and system for inverted Chinese character verification code Download PDFInfo
- Publication number
- CN107292307B CN107292307B CN201710599718.0A CN201710599718A CN107292307B CN 107292307 B CN107292307 B CN 107292307B CN 201710599718 A CN201710599718 A CN 201710599718A CN 107292307 B CN107292307 B CN 107292307B
- Authority
- CN
- China
- Prior art keywords
- image
- character
- verification code
- sample image
- label information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Image Analysis (AREA)
- Character Discrimination (AREA)
Abstract
The invention discloses an automatic identification method and system for an inverted Chinese character verification code, wherein the method comprises the following steps: processing the verification code image to be identified to obtain a target verification code image, and segmenting the target verification code image by the left boundary and the right boundary of each character in the target verification code image to obtain each single-character image; generating label information corresponding to the target verification code image, wherein the label information is represented by binary, 0 represents that characters in the single-character image are normal, and 1 represents that the characters in the single-character image are inverted; the length of the label information is used as a parameter and is transmitted to a trained inverted Chinese character automatic identification model, and the number of output of the neuron is set to be consistent with the length of the label information by the inverted Chinese character automatic identification model; and (4) using the sum of the cross entropies as a loss function of the trained automatic identification model of the inverted Chinese character, training the loss function until the loss function is converged, and outputting an identification result. The automatic identification of the inverted Chinese character verification code can be realized.
Description
Technical Field
the invention belongs to the technical field of automatic identification, and particularly relates to an automatic identification method and system for an inverted Chinese character verification code.
background
with the rapid development of the internet, the network brings great convenience to the life of people, and meanwhile, the security problem of the network is increasingly prominent. The network verification code is a widely used verification means, and plays an important role in network security.
At present, the identification of the character type verification code is mainly the identification of characters, and the main flow is as follows: preparing an original picture material; preprocessing a picture, cutting characters and normalizing the size of the picture; marking picture characters; extracting character and picture features; generating a training data set corresponding to the features and the labels; training the feature tag data to generate a recognition model; a new set of unknown pictures is predicted using the recognition model.
however, the conventional verification code recognition technology is to recognize characters, such as numbers, letters, etc., appearing in the verification code. Without regard to how the inverted chinese character is recognized.
disclosure of Invention
In view of the above defects or improvement requirements of the prior art, the present invention provides an automatic identification method and system for an inverted chinese character verification code, so as to solve the technical problem that the traditional verification code identification technology cannot identify inverted chinese characters.
To achieve the above object, according to one aspect of the present invention, there is provided an automatic identification method of an inverted chinese character verification code, comprising:
processing a verification code image to be recognized to obtain a target verification code image, acquiring the left and right boundaries of each character in the target verification code image, and segmenting the target verification code image according to the left and right boundaries of each character to obtain single-character images;
Generating label information corresponding to the target verification code image, wherein the label information is represented by binary, 0 represents that characters in a single-character image are normal, and 1 represents that characters in the single-character image are inverted;
Transmitting the length of the label information as a parameter to a trained automatic inverted Chinese character recognition model, and setting the number of output of the neuron to be consistent with the length of the label information by the trained automatic inverted Chinese character recognition model, wherein each output is a two-class classifier;
and using the sum of cross entropies as a loss function of the trained automatic inverted Chinese character recognition model, training the loss function until the loss function is converged, and outputting a recognition result, wherein the number of the cross entropies is the same as the number of outputs.
preferably, the training method of the trained automatic inverted Chinese character recognition model comprises the following steps:
Respectively processing each sample image in the test sample image to obtain the left and right boundaries of each character in each processed sample image, and segmenting each sample image by the left and right boundaries of each character in each sample image to obtain a single character image corresponding to each sample image;
Respectively generating label information corresponding to each sample image, wherein the label information of each sample image is represented by binary, 0 represents that characters in a single-character image in the corresponding sample image are normal, and 1 represents that characters in the single-character image in the corresponding sample image are inverted;
Sequentially transmitting the length of the label information of each sample image as a parameter to a training model, and setting the output number of neurons by the training model according to the length of the input label information, wherein each output is a two-classification classifier;
And training the training model by using the sum of the cross entropies as a loss function of the training model until the output accuracy of the training model meets the requirement of preset accuracy, and storing the training model.
Preferably, the training model is a 2-dimensional convolutional neural network model, and the 2-dimensional convolutional neural network model adopts a multilayer convolutional layer plus a multilayer pooling layer, and finally a full-link layer.
Preferably, the parameters of the 2-dimensional convolutional neural network model are that the input is the number of characters in the verification code image, the number of convolutional kernels is M, M-dimensional features are extracted for each input, the size of the convolutional kernels is N × N, the channel value is L, the boundary processing mode padding value is SAME, the pooling function adopts average pooling avg _ pool, and C × C pooling is used, that is, the color block of C × C is reduced to be C × CThe boundary processing mode padding value is SAME.
Preferably, M is 32, N is 5, L is 1 and C is 2.
According to another aspect of the present invention, there is provided an automatic identification system of an inverted chinese character verification code, comprising:
the character segmentation module is used for processing a verification code image to be identified to obtain a target verification code image, acquiring the left and right boundaries of each character in the target verification code image, and segmenting the target verification code image according to the left and right boundaries of each character to obtain single-character images;
the tag information generating module is used for generating tag information corresponding to the target verification code image, wherein the tag information is represented by binary, 0 represents that characters in a single character image are normal, and 1 represents that characters in the single character image are inverted;
The input module is used for transmitting the length of the label information as a parameter to a trained automatic inverted Chinese character recognition model, and the trained automatic inverted Chinese character recognition model sets the number of output of the neuron to be consistent with the length of the label information, wherein each output is a two-class classifier;
and the recognition module is used for training the loss function by using the sum of cross entropies as the loss function of the trained automatic inverted Chinese character recognition model until the loss function is converged and outputting a recognition result, wherein the number of the cross entropies is the same as the number of outputs.
preferably, the character segmentation module is further configured to process each sample image in the test sample image, obtain left and right boundaries of each character in each processed sample image, and segment each sample image by the left and right boundaries of each character in each sample image to obtain a single character image corresponding to each sample image;
The label information generating module is further used for respectively generating label information corresponding to each sample image, wherein the label information of each sample image is represented by binary, 0 represents that characters in a single-character image in the corresponding sample image are normal, and 1 represents that characters in the single-character image in the corresponding sample image are inverted;
The input module is further used for sequentially transmitting the length of the label information of each sample image as a parameter to a training model, and the training model sets the output number of the neurons according to the length of the input label information, wherein each output is a two-class classifier;
and the model training module is used for training the training model by using the sum of cross entropies as a loss function of the training model until the output accuracy of the training model meets the requirement of preset accuracy, and storing the training model.
preferably, the training model is a 2-dimensional convolutional neural network model, and the 2-dimensional convolutional neural network model adopts a multilayer convolutional layer plus a multilayer pooling layer, and finally a full-link layer.
Preferably, the parameters of the 2-dimensional convolutional neural network model are that the input is the number of characters in the verification code image, the number of convolutional kernels is M, M-dimensional features are extracted for each input, the size of the convolutional kernels is N × N, the channel value is L, the boundary processing mode padding value is SAME, the pooling function adopts average pooling avg _ pool, and C × C pooling is used, that is, the color block of C × C is reduced to be C × Cthe boundary processing mode padding value is SAME.
Preferably, M is 32, N is 5, L is 1 and C is 2.
in general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects: the verification code image is segmented to obtain single character images, and then the binary sequence is used for distinguishing inverted characters from normal characters, so that the inverted characters and the normal characters can be accurately distinguished; the length of the generated label information containing character characteristics in the verification code image is used as the input of the trained automatic identification model of the inverted Chinese character, so that the automatic identification of the inverted Chinese character verification code can be realized; the inverted Chinese character automatic identification model is obtained by training the constructed training model through the segmented test sample image, inverted characters in the Chinese character verification code can be automatically identified, manual operation is reduced, and an automatic process is realized.
Drawings
fig. 1 is a schematic flow chart of an automatic identification method for an inverted chinese character verification code according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Fig. 1 is a schematic flow chart of an automatic identification method for an inverted chinese character verification code according to an embodiment of the present invention, where the method shown in fig. 1 includes the following steps:
s1, processing the verification code image to be recognized to obtain a target verification code image, obtaining the left and right boundaries of each character in the target verification code image, and segmenting the target verification code image according to the left and right boundaries of each character to obtain single-character images;
as an alternative embodiment, in step S1, the following method may be adopted to process the verification code image to be recognized:
preprocessing the image of the verification code to be identified, and mainly comprising the technologies of graying, binaryzation, noise suppression and the like.
1. Graying of image
A color value in the RGB system consists of 3 components, such an image is called a color image, and the RGB system is called a color space model. Common color space models are also HSI, CMYK, etc. If the color space of an image is one-dimensional (a color value has only one color component), the image is a grayscale image. In the bitmap image, a grayscale image is generally displayed with R ═ G ═ B.
The following three methods are commonly used for graying:
g(x,y)＝0.11×R(x,y)+0.59×G(x,y)+0.3×B(x,y) (2.2)
g(x,y)＝min(R(x,y),G(x,y),B(x,y)) (2.3)
wherein, the method of formula (2.1) is derived from the formula for calculating the I component in the I color space, and formula (2.2) is derived from the formula for calculating the Y component in the NTSC color space. Equation (2.3) is based on using a method that preserves minimum brightness (black).
An RGB color image can be regarded as being composed of 3 monochromatic grayscale images, and a grayscale image can be obtained by directly taking any one of the RGB channels, such as g (x, y) ═ B (x, y), if the luminance information of the target pixel in the image is mainly distributed on the B channel, otherwise the grayscale result will be a large loss of luminance information. The gray image is also called a brightness image, the brightness is represented by the normalized value, the maximum value represents white, and the minimum value represents black.
let P (x, y) denote a point in the image, x, y are the abscissa and ordinate of the image, respectively, R (x, y) denotes the color component of the R channel, G (x, y) denotes the color component of the G channel, and B (x, y) denotes the color component of the B channel. The luminance value of the point P (x, y) is represented by L (x, y). The brightness of the color image is not strictly defined and calculated, and is generally calculated by the formula (2.1), which is denoted as L1(x, y). The same luminance value calculated by the formula (2.2) is denoted as L2(x, y), and the luminance value calculated by the formula (2.3) is denoted as L3(x, y). It can be demonstrated that:
L3(x,y)≤L1(x,y) (2.4)
L3(x,y)≤L2(x,y) (2.5)
The formula (2.1) takes the average value of RGB channels, the obtained image is relatively soft, and meanwhile, the average brightness difference between the target and the background is reduced, so that the subsequent threshold processing is not facilitated. The formula (2.2) considers that the human eye has the strongest fitness to green, the second blue and the worst red. When processing the green and blue tone verification code image, the effect of equation (2.2) is satisfactory, but when processing the red tone image, because the weight of red in the equation is small, the brightness difference between the target pixel and the background pixel after graying is severely reduced, and the effect is not as good as that of equation (2.1). The formula (2.3) is based on a premise that the brightness information of the target pixel is limited and reserved, and the subsequent threshold segmentation is facilitated.
2. Binarization of images
The gray scale map of a typical 24-bit RGB image is 8-bit 256 gray scales, and if this gray scale is reduced to 1-bit 2 gray scales, a binary image is obtained in which all data are 0 or 1.
As an alternative implementation, after the target verification code image is obtained by processing the verification code image to be recognized, the target verification code image needs to be segmented, where the character segmentation includes segmenting a character region from the verification code image and dividing the character region into two parts of a single character. If statistical feature matching and neural network recognition are used, individual characters must be segmented first. Simple segmentation methods include equidistant segmentation, integral projection segmentation, intersection segmentation, communication area finding, and the like. And cutting the characters into a group of character series pictures to be detected. And outputting a 0,1 sequence with the same number according to the number of the characters in the picture (wherein 0 represents that the Chinese character is normal, and 1 represents that the Chinese character is inverted). And a step of processing the divided picture, namely scanning the left and right boundaries of each character, namely finding out the left and right most points of each character, and cutting according to the boundaries.
S2, generating label information corresponding to the target verification code image, wherein the label information is represented by binary, 0 represents that the characters in the single-character image are normal, and 1 represents that the characters in the single-character image are inverted;
s3, transferring the length of the label information as a parameter to a trained automatic inverted Chinese character recognition model, and setting the number of output of the neuron to be consistent with the length of the label information by the trained automatic inverted Chinese character recognition model, wherein each output is a two-class classifier;
and S4, training the loss function by using the sum of the cross entropies as the loss function of the trained inverted Chinese character automatic recognition model until the loss function is converged, and outputting a recognition result, wherein the number of the cross entropies is the same as the number of outputs.
where the recognition result is similar to 010100 (binary number equals the number of characters, where 1 indicates the kanji inversion).
the training method of the trained automatic identification model of the inverted Chinese characters comprises the following steps:
Respectively processing each sample image in the test sample image to obtain the left and right boundaries of each character in each processed sample image, and segmenting each sample image by the left and right boundaries of each character in each sample image to obtain a single character image corresponding to each sample image;
Respectively generating label information corresponding to each sample image, wherein the label information of each sample image is represented by binary, 0 represents that characters in a single-character image in the corresponding sample image are normal, and 1 represents that characters in the single-character image in the corresponding sample image are inverted;
Sequentially transmitting the length of the label information of each sample image as a parameter to a training model, and setting the output number of the neurons by the training model according to the length of the input label information, wherein each output is a two-classification classifier;
After parameterization is carried out by adopting the method, the method can adapt to the conditions of characters with different numbers, and achieves better universality.
and training the training model by using the sum of the cross entropies as a loss function of the training model until the output accuracy of the training model meets the requirement of preset accuracy, and storing the training model.
Preferably, a tool kit based on TensorFlow and TF-Slim can be adopted to train the training model, the optimizer can select Adam, the accuracy is finally calculated, the model is saved, and the requirement of the preset accuracy can be determined as required.
The training model is a 2-dimensional convolutional neural network model, the 2-dimensional convolutional neural network model adopts a multilayer convolutional layer, a multilayer pooling layer and a full-connection layer, the output is a 0,1 sequence (the total number of 0,1 is the same as the number of single characters in the verification code image), and preferably, the 2-dimensional convolutional neural network model provided by Tensorflow can be used.
The parameters of the 2-dimensional convolutional neural network model are that the input is the number of characters in the verification code image, the number of convolutional kernels is M, M-dimensional features are extracted from each input, the size of the convolutional kernels is N, the channel value is L, the boundary processing mode padding value is SAME, the pooling function adopts average pooling avg _ pool, and C pooling is used, namely C color blocks are reduced to be SAMEThe boundary processing mode padding value is SAME. Preferably, M is 32, N is 5, L is 1 and C is 2.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (8)
1. An automatic identification method for an inverted Chinese character verification code is characterized by comprising the following steps:
processing a verification code image to be recognized to obtain a target verification code image, acquiring the left and right boundaries of each character in the target verification code image, and segmenting the target verification code image according to the left and right boundaries of each character to obtain single-character images;
generating label information corresponding to the target verification code image, wherein the label information is represented by binary, 0 represents that characters in a single-character image are normal, and 1 represents that characters in the single-character image are inverted;
transmitting the length of the label information as a parameter to a trained automatic inverted Chinese character recognition model, and setting the number of output of the neuron to be consistent with the length of the label information by the trained automatic inverted Chinese character recognition model, wherein each output is a two-class classifier; the automatic identification model of the inverted Chinese characters is a 2-dimensional convolutional neural network model, and the 2-dimensional convolutional neural network model adopts a multilayer convolutional layer, a multilayer pooling layer and a full-connection layer;
And using the sum of cross entropies as a loss function of the trained automatic inverted Chinese character recognition model, training the loss function until the loss function is converged, and outputting a recognition result, wherein the number of the cross entropies is the same as the number of outputs.
2. The method of claim 1, wherein the training method of the trained automatic inverted Chinese character recognition model comprises the following steps:
Respectively processing each sample image in the test sample image to obtain the left and right boundaries of each character in each processed sample image, and segmenting each sample image by the left and right boundaries of each character in each sample image to obtain a single character image corresponding to each sample image;
Respectively generating label information corresponding to each sample image, wherein the label information of each sample image is represented by binary, 0 represents that characters in a single-character image in the corresponding sample image are normal, and 1 represents that characters in the single-character image in the corresponding sample image are inverted;
sequentially transmitting the length of the label information of each sample image as a parameter to a training model, and setting the output number of neurons by the training model according to the length of the input label information, wherein each output is a two-classification classifier;
and training the training model by using the sum of the cross entropies as a loss function of the training model until the output accuracy of the training model meets the requirement of preset accuracy, and storing the training model.
3. the method of claim 2, wherein the parameters of the 2-dimensional convolutional neural network model are that the inputs are the number of characters in the captcha image, the number of convolutional kernels is M, M-dimensional features are extracted for each input, the size of the convolutional kernels is N x N, the channel value is L, the boundary processing mode padding value is SAME, the pooling function uses average pooling avg _ pool, C is used for pooling C, and C is used for reducing C color blocks to C color blocksThe boundary processing mode padding value is SAME.
4. The method of claim 3, wherein M is 32, N is 5, L is 1, and C is 2.
5. An automatic identification system for an inverted Chinese character verification code is characterized by comprising:
the character segmentation module is used for processing a verification code image to be identified to obtain a target verification code image, acquiring the left and right boundaries of each character in the target verification code image, and segmenting the target verification code image according to the left and right boundaries of each character to obtain single-character images;
the tag information generating module is used for generating tag information corresponding to the target verification code image, wherein the tag information is represented by binary, 0 represents that characters in a single character image are normal, and 1 represents that characters in the single character image are inverted;
the input module is used for transmitting the length of the label information as a parameter to a trained automatic inverted Chinese character recognition model, and the trained automatic inverted Chinese character recognition model sets the number of output of the neuron to be consistent with the length of the label information, wherein each output is a two-class classifier; the automatic identification model of the inverted Chinese characters is a 2-dimensional convolutional neural network model, and the 2-dimensional convolutional neural network model adopts a multilayer convolutional layer, a multilayer pooling layer and a full-connection layer;
And the recognition module is used for training the loss function by using the sum of cross entropies as the loss function of the trained automatic inverted Chinese character recognition model until the loss function is converged and outputting a recognition result, wherein the number of the cross entropies is the same as the number of outputs.
6. the system of claim 5, wherein the character segmentation module is further configured to process each sample image in the test sample image, to obtain left and right boundaries of each character in each processed sample image, and to segment each sample image by the left and right boundaries of each character in each sample image to obtain a single character image corresponding to each sample image;
The label information generating module is further used for respectively generating label information corresponding to each sample image, wherein the label information of each sample image is represented by binary, 0 represents that characters in a single-character image in the corresponding sample image are normal, and 1 represents that characters in the single-character image in the corresponding sample image are inverted;
The input module is further used for sequentially transmitting the length of the label information of each sample image as a parameter to a training model, and the training model sets the output number of the neurons according to the length of the input label information, wherein each output is a two-class classifier;
and the model training module is used for training the training model by using the sum of cross entropies as a loss function of the training model until the output accuracy of the training model meets the requirement of preset accuracy, and storing the training model.
7. The system of claim 6, wherein the parameters of the 2-dimensional convolutional neural network model are that the inputs are the number of characters in the captcha image, the number of convolutional kernels is M, M-dimensional features are extracted for each input, the size of the convolutional kernels is N x N, the channel value is L, the boundary processing mode padding value is SAME, the pooling function uses average pooling avg _ pool, and C pooling is used, i.e., C color blocks are reduced to C color blocksThe boundary processing mode padding value is SAME.
8. the system of claim 7, wherein M is 32, N is 5, L is 1, and C is 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710599718.0A CN107292307B (en) | 2017-07-21 | 2017-07-21 | Automatic identification method and system for inverted Chinese character verification code |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710599718.0A CN107292307B (en) | 2017-07-21 | 2017-07-21 | Automatic identification method and system for inverted Chinese character verification code |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107292307A CN107292307A (en) | 2017-10-24 |
| CN107292307B true CN107292307B (en) | 2019-12-17 |
Family
ID=60102053
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710599718.0A Expired - Fee Related CN107292307B (en) | 2017-07-21 | 2017-07-21 | Automatic identification method and system for inverted Chinese character verification code |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107292307B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107992527B (en) * | 2017-11-09 | 2021-01-29 | 武汉极意网络科技有限公司 | Data mark checking method, server and storage medium |
| CN107886128A (en) * | 2017-11-10 | 2018-04-06 | 广东工业大学 | A kind of shuttlecock recognition methods, system, medium and equipment |
| CN107967475B (en) * | 2017-11-16 | 2020-04-14 | 广州探迹科技有限公司 | Verification code identification method based on window sliding and convolutional neural network |
| CN108537115B (en) * | 2018-03-02 | 2022-01-25 | 创新先进技术有限公司 | Image recognition method and device and electronic equipment |
| CN108572593B (en) * | 2018-04-27 | 2020-12-18 | 北京源码矩阵科技有限公司 | Cross-platform convolutional neural network control system and method and information data processing terminal |
| CN108764242A (en) * | 2018-05-21 | 2018-11-06 | 浙江工业大学 | Off-line Chinese Character discrimination body recognition methods based on deep layer convolutional neural networks |
| CN111753575A (en) * | 2019-03-26 | 2020-10-09 | 杭州海康威视数字技术股份有限公司 | Text recognition method, device and equipment |
| CN111178162B (en) * | 2019-12-12 | 2023-11-07 | 北京迈格威科技有限公司 | Image recognition method, device, computer equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102419817A (en) * | 2010-09-27 | 2012-04-18 | 贵州黔驰电力信息技术有限公司 | Automatic document scanning, analyzing and processing system based on intelligent image identification |
| CN105335754A (en) * | 2015-10-29 | 2016-02-17 | 小米科技有限责任公司 | Character recognition method and device |
| CN106096602A (en) * | 2016-06-21 | 2016-11-09 | 苏州大学 | A kind of Chinese licence plate recognition method based on convolutional neural networks |
| CN106250756A (en) * | 2016-07-29 | 2016-12-21 | 智者四海（北京）技术有限公司 | Generation method, verification method and the related device of identifying code |
| CN106446897A (en) * | 2016-09-09 | 2017-02-22 | 浪潮软件股份有限公司 | Hollow verification code identification method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8774558B2 (en) * | 2010-11-29 | 2014-07-08 | Microsoft Corporation | Rectification of characters and text as transform invariant low-rank textures |
-
2017
- 2017-07-21 CN CN201710599718.0A patent/CN107292307B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102419817A (en) * | 2010-09-27 | 2012-04-18 | 贵州黔驰电力信息技术有限公司 | Automatic document scanning, analyzing and processing system based on intelligent image identification |
| CN105335754A (en) * | 2015-10-29 | 2016-02-17 | 小米科技有限责任公司 | Character recognition method and device |
| CN106096602A (en) * | 2016-06-21 | 2016-11-09 | 苏州大学 | A kind of Chinese licence plate recognition method based on convolutional neural networks |
| CN106250756A (en) * | 2016-07-29 | 2016-12-21 | 智者四海（北京）技术有限公司 | Generation method, verification method and the related device of identifying code |
| CN106446897A (en) * | 2016-09-09 | 2017-02-22 | 浪潮软件股份有限公司 | Hollow verification code identification method |
Non-Patent Citations (7)
| Title |
|---|
| A Low-cost Attack on a Microsoft CAPTCHA;Jeff Yan 等;《School of Computing Science》;20081231;第1-19页 * |
| A New Segmentation Method for Connected Characters in CAPTCHA;Pengpeng Lu 等;《2015 International Conference on Control, Automation and Information Sciences》;20151031;第128-131页 * |
| 中文文本图像倒置快速检测算法;曾凡锋 等;《计算机工程与设计》;20120930;第33卷(第9期);第3512-3516页 * |
| 卷积神经网络在验证码识别上的应用与研究;刘欢 等;《计算机工程与应用》;20161231;第52卷(第18期);第1-7页 * |
| 基于卷积神经网络的铁路货运网站验证码识别;陈超 等;《指挥信息***与技术》;20160831;第7卷(第4期);第91-96页 * |
| 基于汉字笔画特征的文本图像倒置判断算法;王景中 等;《计算机技术与发展》;20140531;第24卷(第5期);第129-133页 * |
| 基于神经网络的数字验证码识别研究;吕刚 等;《浙江工业大学学报》;20100831;第38卷(第4期);第433-436页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107292307A (en) | 2017-10-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107292307B (en) | Automatic identification method and system for inverted Chinese character verification code | |
| CN111401372B (en) | Method for extracting and identifying image-text information of scanned document | |
| CN107622274B (en) | Neural network training method and device for image processing and computer equipment | |
| CN106815560B (en) | Face recognition method applied to self-adaptive driving seat | |
| CN108280426B (en) | Dark light source expression identification method and device based on transfer learning | |
| CN106127817B (en) | A kind of image binaryzation method based on channel | |
| CN109740572A (en) | A kind of human face in-vivo detection method based on partial color textural characteristics | |
| CN109145964B (en) | Method and system for realizing image color clustering | |
| CN110930296A (en) | Image processing method, device, equipment and storage medium | |
| CN103530625A (en) | Optical character recognition method based on digital image processing | |
| CN108647696B (en) | Picture color value determining method and device, electronic equipment and storage medium | |
| CN110991434B (en) | Self-service terminal certificate identification method and device | |
| CN112633221A (en) | Face direction detection method and related device | |
| US7620246B2 (en) | Method and apparatus for image processing | |
| CN113723410B (en) | Digital identification method and device for nixie tube | |
| US10764471B1 (en) | Customized grayscale conversion in color form processing for text recognition in OCR | |
| CN109657544B (en) | Face detection method and device | |
| CN108491820B (en) | Method, device and equipment for identifying limb representation information in image and storage medium | |
| JPH11306325A (en) | Method and device for object detection | |
| CN109766860A (en) | Method for detecting human face based on improved Adaboost algorithm | |
| KR20030091471A (en) | YCrCb color based human face location detection method | |
| CN111754459B (en) | Dyeing fake image detection method based on statistical depth characteristics and electronic device | |
| CN110348530B (en) | Method for identifying lipstick number | |
| CN108764106B (en) | Multi-scale color image face comparison method based on cascade structure | |
| CN111242047A (en) | Image processing method and apparatus, electronic device, and computer-readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191217 Termination date: 20200721 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |